Predict returning NaN

[John Mann]

I've been trying to put together a simple example of Tensorflow.js and appear to have the code below all working apart from when I do the final predict, when I always get NaN returned.

What I was looking for was some sort of example where data when into a model and some predictions came out, and I couldn't really find anything suitable from the Tensorflow.js examples, however the code in this video seemed just what I wanted:

https://www.youtube.com/watch?v=Y_XM3Bu-4yc

This is training the model where simpleInput and simpleOutput are just one dimensional arrays.

    function ML_RunTensor2(simpleInput, simpleOutput) {

        try {

            // define a model for linear regression

            linearModel = tf.sequential();

            linearModel.add(tf.layers.dense({ units: 1, inputShape: [1] }));

            // prepare the model for training: specify the loss and the optimizer

            linearModel.compile({ loss: 'meanSquaredError', optimizer: 'sgd' });

            // training data

            const xs = tf.tensor1d(simpleInput);

            const ys = tf.tensor1d(simpleOutput);

            // train

            linearModel.fit(xs, ys);

        } catch (e) {

            debugger;

            console.log(e.Message);

        }

    }

Then ones the model is trained I call this for a prediction:

    function ML_PredictClicked() {

        var inputPrediction = jQuery('#' + predictionInput).val();

        var inputPredictionNumber = parseFloat(inputPrediction);

        const output = linearModel.predict(tf.tensor2d([inputPredictionNumber], [1, 1]));

        var outputPrediction = Array.from(output.dataSync())[0];      

        jQuery('#' + predictionOutput).val(outputPrediction);

    }

Here is the code in the debugger - I tried just output.dataSync() as well.

[Nikhil Thorat]

Hi John,

If you call "xs.dataSync()" and "ys.dataSync()" are you getting any NaN values?

--
You received this message because you are subscribed to the Google Groups "TensorFlow.js Discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tfjs+unsubscribe@tensorflow.org.
Visit this group at https://groups.google.com/a/tensorflow.org/group/tfjs/.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/tfjs/7b78feea-da45-4f33-9c55-d4bb2738a229%40tensorflow.org.

[John Mann]

Hi Nikhil thanks for replying. Yes I am - I can see now what the problem is - the simpleInputs array turned out to be objects not numbers. The simplyOutputs was fine. THANK YOU.

To unsubscribe from this group and stop receiving emails from it, send an email to tfjs+uns...@tensorflow.org.